Re: Q: perf log mode?

[Frederic Weisbecker <fweisbec@gmail.com>]

On Wed, Jan 12, 2011 at 03:53:41PM +0100, Peter Zijlstra wrote:
> On Wed, 2011-01-12 at 15:42 +0100, Frederic Weisbecker wrote:
> > On Wed, Jan 12, 2011 at 03:08:08PM +0100, Peter Zijlstra wrote:
> > > On Wed, 2011-01-12 at 17:06 +0300, Kirill Smelkov wrote:
> > > > I'm trying to use perf together with e.g. kprobes as a tool to show what
> > > > is happening with my system in "live-log" mode. The problem is, for
> > > > seldom events, actual info output is largely delayed because perf reads
> > > > sample data in whole pages. Could something be done with it or am I'm
> > > > missing something? Here is detailed description: 
> > > 
> > > perf_event_attr = {
> > > 	.watermark = 0,
> > > 	.wakeup_events = 1,
> > > };
> > 
> > 
> > Which is perhaps something we want as a default when perf record -c 1
> > and the output is the pipe mode.
> 
> No, definitely not, esp for -c1 you want large buffers because the event
> can come at very high freq.

Right this should not be a default actually if we have high freq events
we don't want wake up at each of them.

> Nor does pipe mode have anything to do with it, the whole script set-up
> plain stinks 

Oh? What's wrong with the scripts infrastructure?

> and should not be using pipe mode, pipe mode should only be
> used to pipe data over the network and other remote profiling like
> things. 

But the scripts are just an endpoint over many other possibilities,
regardless of what is in the middle: perf.data or pipe streaming.

Where is a limitation on using scripts here?